Age-matched dumps of 126 Stack Exchange sites show 67% fewer questions in 2025 than in 2022, a larger decrease in technology sites, and no clear change in the 7-day answer rate

Question: how did question and answer activity change across the Stack Exchange network after the release of ChatGPT (30 November 2022), and where was the decrease largest? Approach: we extracted all main sites (not Stack Overflow) from seven public data dumps (2021-12 to 2026-09). A single dump does not contain deleted posts, and we found that many young questions are deleted after approximately one year (median survival 0.905 for 2023 questions from the 2023 to the 2025 dump; 0.83 in technology sites). Thus, we measured each year in its own year-end dump, so that all years have the same age and deletion exposure. Results for 126 old, active sites: the median site had 0.33 times its 2022 question count in 2025 (log change -1.11, 95% CI -1.21 to -1.04), and approximately 0.13 times in January-August 2026. Technology sites decreased more than other sites (adjusted log difference -0.52, CI -0.74 to -0.30; 2021-2022 placebo +0.01). Sites with more code in their questions decreased more, but this relation was not resolved inside the technology category, and code and non-code questions in the same site decreased equally. The share of questions answered within 7 days did not change (-0.004, CI -0.030 to +0.007), but time to first answer increased from 3.7 to 5.7 hours. Answerers decreased less than questions. A single-dump analysis gives a false decrease in answer coverage. Limitations: observational design; other 2023-2025 events (policy changes, moderation, search changes) can contribute; Stack Overflow itself is only compared descriptively.

Stack Exchange · online communities · generative AI · ChatGPT · question answering · collective knowledge production · data deletion bias · survivorship bias · difference in differences · knowledge commons